Keelin Murphy, Presenter: Nothing to Disclose

Mathias Prokop MD, Abstract Co-Author: Nothing to Disclose

Cornelia Maria Schaefer-Prokop MD, Abstract Co-Author: Nothing to Disclose

Hester Gietema MD, Abstract Co-Author: Nothing to Disclose

G. D. Nossent MD, PhD, Abstract Co-Author: Nothing to Disclose

Bram Van Ginneken PhD, Abstract Co-Author: Nothing to Disclose

Josien Pluim, Abstract Co-Author: Nothing to Disclose

Yulia Arzhaeva, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

To evaluate the effect of presenting accurately registered images on the speed and accuracy of observers in assessing progression of interstitial lung disease.

24 patients with ILD were included with baseline and follow-up CT scans (Philips Brilliance 16, 16x0.75mm collimation, 50-120mAs, voxel sizes ~0.4 mm). Scans were presented in pairs to 4 independent observers. Observers comprised 2 experienced thoracic radiologists, 1 resident and 1 pulmonologist. Each observer saw every scan-pair twice, in separate sessions, A and B. In session A each scan-pair was presented side-by-side randomly in either a “registered view” (RV) or a “non-registered view” (NRV). The RV depicted one scan in its original form while the other scan was shown after automatic elastic registration to align it anatomically with its pair. The registration used a non-rigid transformation modelled using B-splines and a mutual-information cost function. The NRV depicted both scans in their original form. In session B the observers assessed all scan pairs again and the view-type (RV or NRV) for each pair was opposite to that used in session A. At each viewing the observer was asked to say which scan displayed more ILD or if there was no change. The 2 most experienced experts provided a consensus ground-truth which they later obtained by examining the scan-pairs together in both RV and NRV views.

Assessment speed was reduced from an average of 123 seconds per scan-pair in NRV to an average of 79 seconds in RV. The speed of assessment was reduced for every observer individually and for every category of ILD progression. Interobserver variability was given by (Fleiss’ kappa) κ=0.54 for NRV and κ=0.58 for RV. Agreement with the ground-truth score was measured at an average weighted (Cohen’s kappa) κ=0.66 for NRV and κ=0.67 for RV.

Assessment of ILD progression was substantially faster (36% time reduction) when viewing registered images without affecting accuracy or interobserver variability.

Automatic elastic registration of temporal chest CT scans with ILD is feasible and greatly reduces reading time without affecting accuracy of interval change assessment.

Murphy, K, Prokop, M, Schaefer-Prokop, C, Gietema, H, Nossent, G, Van Ginneken, B, Pluim, J, Arzhaeva, Y, et al, , Improved Efficiency of Assessment of Interstitial Lung Disease Progression in HRCT of the Chest by Visualisation of Automatically-registered Image Pairs.  Radiological Society of North America 2008 Scientific Assembly and Annual Meeting, February 18 - February 20, 2008 ,Chicago IL. http://archive.rsna.org/2008/6020368.html